the main class of gml

class GML

The main process of progressive machine learning, which runs through Essential Support, Approximate Probability Estimation, select topm, select topk, construct subgraph, inference subgraph process

This class currently provides the following methods: 

1. evidential_support(variable_set,update_feature_set)

Function: calculate essential support
parameter:
variable_set - the set of target variables
update_feature_set - the set of target features
Return: none
Return type：none 

2. approximate_probability_estimation(variable_set)

Function: Calculate approximate probability
parameter:
variable_set - the set of target variables
Return: none
Return type：none 

3. select_top_m_by_es(m)

Function: Select the first m latent variables according to the calculated Evidential Support (large to small) parameter:
m - The number of latent variables to be selected
Return: a list containing m variable ids
Return type：list

4. select_top_k_by_entropy(var_id_list, k)

Function: calculate entropy, select top_k latent variables with small entropy
parameter:
var_id_list - Choose range
· k - The number of latent variables to be selected
Return: a list containing k ids
Return type：list

5. select_evidence(var_id)

Function: Select the edges, variables and features which needed for subsequent subgraph construction
parameter:
var_id - The id of the target variable
Return: Edges, variables and features needed for subsequent subgraph construction
Return type：set

6. construct_subgraph(var_id)

Function: Select topk latnet variables and create subgraph
parameter:
var_id - The id of the target variable
Return: According to the factor graph requirement of numbskull,return weight, variable, factor, fmap, domain_mask, edges
Return type: multiple types

7. inference_subgraph(var_id)

Function: inference subgraph
parameter:
var_id - For entity recognition, var_id is a variable id, and var_id is a set of k variables for sentiment analysis
Return: none
Return type：none 

8. label(var_id_list)

Function: Compare the entropy of k latent variables, select the one with the smallest entropy and label it, and write the parameters learned from this graph back to self.features parameter:
var_id_list - A list of k ids, the probability corresponding to each variable is taken from variables
Return: No output, directly update the label and entropy in vairables, by the way, you can update observed_variables_id and potential_variables_id
Return type：dict

9. inference()

Function: Main flow
parameter:
none
Return: none
Return type：none 

10. score() [[source]]

Function: Calculate the accuracy rate, precision rate, recall rate, f1 value of inference results, etc.
parameter:
none
Return: none
Return type：none 

11. save_results()

Function: Save the variable and feature information after inference.
parameter:
none
Return: none
Return type：none

the class for Approximate Probability Estimation：

class ApproximateProbabilityEstimation(variables, features, method)
Methods for Approximate Probability Estimation class

Parameters:
variables -variables
features -features
method -The method used to calculate the approximate probability

This class currently provides the following methods: 

1. init_binary_feature_weight(value1,value2)

Function：Set initial value of binary feature weight
Parameters:
value1 - binary feature initial weight value 1
value2 - binary feature initial weight value 2
Returns：binary feature weight initial value
Return type：dictionary

2. labeling_conflict_with_ds(mass_functions)

Function：based on (D-S) theory Evidence support measurement
Parameters：
mass_functions - a dict for d-s Returns：Evidence support value
Return type：float

3. get_pos_prob_based_relation(var_id, weight)

Function：Calculate the proportion of positive instances in marked instances with a feature
Parameters：
var_id- Target variable id
weight - feature weight
Returns：Proportion of positive instances in marked instances with a feature
Return type：float 

4. construct_mass_function_for_confict(uncertain_degree, pos_prob, neg_prob)

Function：Evidence support for calculating each feature connected to an unlabeled variable
Parameters：
uncertain_degree - Uncertainty of a feature
pos_prob - Proportion of positive instances in marked instances with a feature
neg_prob - Proportion of negative instances in marked instances with a feature
Returns：MassFunction function
Return type：function

5. construct_mass_function_for_ER(alpha,tau,featureValue) )

Function：Evidence support for calculating each feature connected to an unlabeled variable
Parameters：
theta - Uncertainty of a feature
alpha - Parameters after unaryfactor linear regression
tau - Parameters after unaryfactor linear regression
featureValue - featurevalue
Returns：MassFunction function
Return type：function

6. approximate_probability_estimation()(variable_set)

Function：Calculate the approximate probability of the selected topm hidden variables, used to select topk Parameters：
variable_set Latent variable data set Return: none Return type: none

7. approximate_probability_estimation_by_custom(variable_set)

Function：Calculate the approximate probability of the selected topm hidden variables, used to select topk, user-defined calculation rules
Parameters：
variable_set - Latent variable data set Return: none
Return type: none

the class for Subgraph construction:

class ConstructSbugraph(variables, features, balance)
Factor graph construction class

Parameters：
- variables – input data of gml, the variable of the factor graph.
- features – input data of gml, The characteristics of the factor graph.
- balance – Whether balance the labels when constructing the subgraph. If the number of evidence variable labels(the number of 0 and 1)is quite large, it is set to True, otherwise it is set to False.

This class currently provides the following methods:

1. construct_subgraph(evidences)

Function：A unified construct subgraph method,it is necessary to construct a factor graph containing uary factors(parameterize or not) and binary factors, and this function is called at this time.
Parameters：
evidences - Variables, edges, and feature sets required when constructing factor graphs
Returns：the data for Numbskull inference（weight, variable, factor, fmap, domain_mask, edges_num, var_map, alpha_bound, tau_bound, weight_map_feature, sample_list, wmap, wfactor）
Return type：Multiple types

the class for selecting evidence

class EvidenceSelect(variables,features,balance,interval_evidence_count=10,subgraph_limit_num,k_hop)
the class for selecting evidence, select evidential variables for latent variables, including two methods for selecting evidential variables, and allow users to define their own methods Parameters:
- variables – One of the input data of gml, the variable of the factor graph
- features – One of the input data of gml, the feature of factor graph
- interval_evidence_limit – When dividing interval sampling evidence, the number of evidences sampled in each interval
- subgraph_limit_num – Maximum number of variables allowed in the subgraph
- each_feature_evidence_limit – When sampling randomly, the number of evidence for each single factor sampling

This class currently provides the following methods：

1. evidence_select(var_id)

Function: Provide a unified evidence selection method, which can be used to construct factor graphs containing parameterized single factor, non-parameterized single factor, and double factor. In this case, call this function.
Parameters：
var_id - the id of latent variable
Return：connected_var_set, connected_edge_set, connected_feature_set
Return type：set

the class for computing evidential support

class EvidentialSupport(variables,features,method,evidence_interval_count,interval_evidence_count)
the set of computing evidential support

Parameters：
- variables - variables
- features - factors
- method - The method of computing evidential support, the default value is 'regression'
- evidence_interval_count - The number of intervals, the default value is 10
- interval_evidence_count - The number of variables in each interval, the default value is 200

This class currently provides the following methods： 

1. get_unlabeled_var_feature_evi()

Function: Calculate the ratio of 0 to 1 in the evidential variables associated with each unary feature of each latent variable, and the variable id at the other end of the binary feature
Return : none
Return type : none 

2. separate_feature_value()

Function: Select the easy feature value of each feature for linear regression
Return : none
Return type : none 

3. influence_modeling()

Function: Perform linear regression on the updated feature, save all the results of the regression back to the feature, the key is 'regression'
Parameter：
update_feature_set - the set of updated feature
Return : none
Return type : none 

4. init_tau_and_alpha()

Function: calculate tau and alpha for feature
Parameter：
feature_set - the set of feature_id
Return : none
Return type : none 

5. computer_unary_feature_es(variable_set)

Function: calculate the essential support of each latent variable in a given set of latent variables, suitable for Aspect-level sentiment analysis
Parameter：
variable_set - a set of latent variables 

6. evidential_support(update_feature_set,variable_set)

Function: Calculate the Evidential Support of all latent variables
Parameter：
update_feature_set - the set of updated feature
variable_set -The set of variables to be calculated the evidence support.
Return : none
Return type : none 

7. get_dict_rel_acc()

Function: Calculate the accuracy of different types of relationships
Return : accuracy of different types of relations
Return type : dict 

8. construct_mass_function_for_propensity(uncertain_degree,label_prob,unlabel_prob)

Function: Build a mass function for Evidential Support calculation
Parameter：
uncertain_degree - Feature uncertainty
label_prob - The probability of label matching, which represents the proportion of positive instances for word features and the accuracy of relationship features for relationship features
unlabel_prob - The probability of label unmatch, which represents the proportion of negative instances for word features and 1 minus the accuracy of relationship features for relationship features
Return : MassFunction
Return type : function 

9. construct_mass_function_for_ER(theta)

Function: Build a mass function for Evidential Support calculation
Parameter：
theta - Feature uncertainty
Return : MassFunction
Return type : function 

10. labeling_propensity_with_ds(massfunctions)

Function: Combine different methods for different types of evidences for aspect-level sentiment analysis Parameter：
massfunctions -a variable corresponding massfunctions
Return: Summarized the calculated evidential value
Return type: dict 

11. evidential_support_by_custom()

Function: User-defined method for calculating essential support
Parameter：
variable_set - a set of latent variables Return : none
Return type : none

class Regression(each_feature_easys, n_job, effective_training_count_threshold) The class of related Linear regression, perform linear regression on all features, used for the essential support calculation of Entity Resolution

Parameter：
each_feature_easys - The feature_value of the easy variable owned by each feature
n_job - calculate the number of threads
effective_training_count_threshold- The minimum number of effective samples, the default value is 2

This class currently provides the following methods:

1. perform()

Function: Perform linear regression method, suitable for uary factor

gml toolkit

gml_utils

gml commonly used tools, currently mainly provides the following methods: 

1. load_easy_instance_from_file(filename)

Function: load easy from csv file
Parameter:
filename - the path of csv file
Return: simple instance list
Return type: list

2. separate_variables(variables)

Function: Divide variables into evidential variables and latent variables
Parameter:
variables - variables
Return: evidential variables list, latent variables list
Return type: list

3. init_evidence_interval(evidence_interval_count)

Function: Initialize the evidence interval
Parameter:
evidence_interval_count - the number of intervals
Return: a list containing evidence_interval_count intervals
Return type: list

4. init_evidence(features,evidence_interval,observed_variables_set)

Function: Add the evidence_interval attribute and the evidence_count attribute for each feature
Parameter:
features - features
evidence_interval - evidential interval
observed_variables_set - the set of evidential varibles
Return: none
Return type: none 

5. update_evidence(variables,features,var_id_list,evidence_interval)

Function: update evidence_interval and evidence_count after label variables
Parameter: variables - variables
features- features
var_id_list - Variable collection
evidence_interval - evidence_interval
Return: none
Return type: none 

6. init_bound(variables,features)

Function: init para bound
Parameter: variables - variables
features - features
Return: none
Return type: none 

7. update_bound(variables,features,var_id_list))

Function: update tau and alpha bound after label variables
Parameter: variables - variables
features - features
var_id_list- Variable collection
Return: none
Return type: none 

8. entropy(probability)

Function: calculate entropy after given probability
Parameter:
probability - One probability or list of probabilities
Return: One entropy or a list of entropies
Return type: floating or list 

9.  open_p(weight)

Function: Calculate approximate probability Parameter:
weight - Weight
Return: One entropy or a list of entropies
Return type: floating

10. combine_evidences_with_ds(mass_functions, normalization)

Function: Summarize and calculate the evidential value of different sources Parameter:
mass_functions - mass function
normalization - Whether to normalize
Return: Summarized the calculated evidential value
Return type: list

class Logger(object) The log class is used to output the results to the file and the console at the same time. Currently, the following methods are mainly provided

Parameter:
object – File object 

1. write(message)

Function: Write to file and console at the same time
Parameter:
message - What's written
Return: none
Return type: none